Fault diagnosis method and system based on robust transfer subspace learning

Abstract

The application relates to the technical field of fault diagnosis, and discloses a fault diagnosis method and system based on robust transfer subspace learning. The method comprises the following steps: converting a one-dimensional vibration signal into a two-dimensional image; providing a neural network, and extracting features from the two-dimensional image through the neural network; performing transfer learning according to the extracted features based on a robust transfer subspace learning algorithm; and determining a fault type corresponding to the one-dimensional vibration signal according to a result of the transfer learning. The application can effectively improve the robustness of a system in the process of fault diagnosis, and has low training cost and high stability.

Description

Technical Field

[0001] This invention generally relates to the field of fault diagnosis technology. Specifically, this invention relates to a fault diagnosis method and system based on robust transfer subspace learning. Background Technology

[0002] Machine learning has been widely used in the field of fault diagnosis technology. Traditional machine learning-based fault diagnosis methods usually include fault diagnosis methods based on vibration signal analysis and fault diagnosis methods based on neural network learning.

[0003] Fault diagnosis methods based on vibration signal analysis typically require manual extraction of vibration features based on human experience, combined with signal processing techniques (such as time-frequency analysis and sparse representation). This approach is cumbersome and suffers from low accuracy in manually extracted vibration features. Furthermore, as the data volume increases, traditional signal processing techniques suffer from excessively long processing times for signal points and low utilization rates of these points.

[0004] Fault diagnosis methods based on neural network learning can automatically learn the abstract representation features of raw data, partially solving the problems existing in the aforementioned fault diagnosis methods based on vibration signal analysis. However, current neural network learning typically uses one-dimensional vibration signals as input during training, which results in significant time consumption when inputting signal points into the neural network for training. Furthermore, the neural networks trained using this method exhibit poor transferability, only capable of detecting under single working conditions and difficult to apply to complex working conditions. Summary of the Invention

[0005] To at least partially address the aforementioned problems in the prior art, this invention proposes a fault diagnosis method based on robust transfer subspace learning, comprising the following steps:

[0006] Convert one-dimensional vibration signals into two-dimensional images;

[0007] A neural network is provided, and features are extracted from a two-dimensional image through the neural network;

[0008] A robust transfer subspace learning algorithm is used to perform transfer learning based on extracted features; and

[0009] The fault type corresponding to the one-dimensional vibration signal is determined based on the results of transfer learning.

[0010] In one embodiment of the present invention, converting a one-dimensional vibration signal into a two-dimensional image includes:

[0011] A two-dimensional image matrix is ​​generated from a one-dimensional vibration signal, expressed as follows:

[0012]

[0013] Where P(j,k) represents the pixel of the two-dimensional image matrix, j and k represent the row and column of the image matrix respectively, L represents the signal point segment, L((j-1)×M+k) represents the one-dimensional vibration signal, and ρ represents the image gray value range, wherein the image gray value range ρ can be 0-255.

[0014] In one embodiment of the present invention, converting a one-dimensional vibration signal into a two-dimensional image includes:

[0015] The one-dimensional source domain vibration signal S1 is transformed into a two-dimensional source domain image X1, where the source domain label Y1 of the two-dimensional source domain image X1 is known; and

[0016] The one-dimensional target domain vibration signal S2 is transformed into a two-dimensional target domain image X2.

[0017] In one embodiment of the present invention, a neural network is provided, and extracting features from a two-dimensional image using the neural network includes:

[0018] The neural network is trained using the two-dimensional source domain image X1 as input;

[0019] The source domain feature matrix X is extracted from the two-dimensional source domain image X1 using a trained neural network. s Where s represents the number of two-dimensional source domain images; and

[0020] The target domain feature matrix X is extracted from the two-dimensional target domain image X2 using a trained neural network. t , where t represents the number of two-dimensional target domain images.

[0021] In one embodiment of the present invention, the neural network includes a convolutional layer, a pooling layer, a fully connected layer, and a normalized exponential layer.

[0022] In one embodiment of the present invention, the transfer learning based on the extracted features according to the robust transfer subspace learning algorithm includes:

[0023] Based on the robust transfer subspace learning algorithm, the source domain feature matrix X is... s and the target domain feature matrix X t The transformation is applied to a common subspace, where the transformation matrix P is generated iteratively. During the transformation process, Gaussian and non-Gaussian noise are removed, and the degrees of freedom of the common subspace are increased using a relaxation regression algorithm, as expressed in the following equation:

[0024]

[0025] in, The function represents the relaxation regression algorithm, R represents the learned label matrix, Z represents the reconstruction matrix, E represents the non-Gaussian noise matrix, N represents the Gaussian noise matrix, and ||·|| * The nuclear norm is represented by ||·||1, which represents the 1-norm. F Let F be the norm, and α, β, γ represent the nonnegativity parameters. i This represents the index of the true label of the i-th sample in the source domain;

[0026] The transformation matrix P and the source domain characteristic matrix X are compared. s Multiplication generates a common subspace source domain matrix M1; and

[0027] The transformation matrix P and the target domain feature matrix X are compared. t Multiply to generate the common subspace target domain matrix M2.

[0028] In one embodiment of the present invention, determining the fault type corresponding to the one-dimensional vibration signal based on the result of transfer learning includes:

[0029] The KNN classifier is trained using the source domain matrix M1 of the common subspace and the source domain label Y1 as input; and

[0030] After training, the KNN classifier determines the fault type corresponding to the one-dimensional target domain vibration signal S2 based on the common subspace target domain matrix M2.

[0031] In this context, the term "KNN classifier" refers to a statistical classifier based on the K-Nearest Neighbor Algorithm.

[0032] This invention also proposes a fault diagnosis system based on robust transfer subspace learning, comprising:

[0033] An image conversion module is configured to convert a one-dimensional vibration signal into a two-dimensional image;

[0034] A feature extraction module is configured to train a neural network and extract features from a two-dimensional image using the neural network.

[0035] The transfer learning module is configured to perform transfer learning based on extracted features using a robust transfer subspace learning algorithm; and

[0036] The fault prediction module is configured to determine the fault type corresponding to the one-dimensional vibration signal based on the results of transfer learning.

[0037] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, the computer program performing the steps according to the method when executed by a processor.

[0038] The present invention also proposes a computer system comprising:

[0039] A processor, configured to execute machine-executable instructions; and

[0040] A memory having machine-executable instructions stored thereon, which, when executed by a processor, perform the steps according to the method.

[0041] This invention offers at least the following advantages: It proposes a fault diagnosis method and system based on robust transfer subspace learning, which effectively improves system robustness and offers low training cost and high stability when applied to the field of fault diagnosis. Specifically, it proposes a technique for two-dimensional imaging of vibration signals, storing the features of hundreds of thousands of signal points within the image for analysis. Addressing the low accuracy of manual feature extraction, a convolutional neural network is used to extract features from the two-dimensional image signal, ensuring effective and accurate feature extraction. To address the issue of large convolutional layers and excessive parameters in neural network models, the VGG-19 network model can be used to simplify the network training process and improve training efficiency. To address the high cost and time consumption of neural network training, a network pre-training strategy is employed, training the network model in advance on the dataset and fine-tuning it during application. To address the poor transferability of a single model in cross-condition diagnosis, this invention uses a strategy of network pre-training combined with transfer learning, resulting in good transferability of the algorithm. To address the poor model stability in cross-condition detection, this invention uses a subspace learning algorithm to transfer different data into the same subspace, completing data reconstruction and reducing data differences under different conditions. To address the issue that data reconstruction can disrupt the local and global structure of labels, we apply a 1-norm and a kernel norm to the reconstruction matrix to ensure structural stability. To address the problem that noise in the data can degrade transfer performance, this invention reduces the impact of noise by adding Gaussian and non-Gaussian noise matrices during the transformation process. To address the low degrees of freedom in subspace transformation algorithms, this invention improves the regression algorithm by learning labels from samples, thus providing greater degrees of freedom for the subspace. Attached Figure Description

[0042] To further illustrate the advantages and other features of the various embodiments of the present invention, a more specific description of the embodiments of the present invention will be presented with reference to the accompanying drawings. It is understood that these drawings depict only typical embodiments of the invention and are therefore not intended to limit its scope. In the drawings, identical or corresponding parts will be indicated by the same or similar reference numerals for clarity.

[0043] Figure 1 A computer system implementing the system and / or method according to the present invention is shown.

[0044] Figure 2 The diagram shows a flowchart of a fault diagnosis method based on robust transfer subspace learning in one embodiment of the present invention.

[0045] Figure 3 The diagram illustrates a process for converting a one-dimensional signal into a two-dimensional image according to one embodiment of the present invention.

[0046] Figure 4 A schematic diagram of image feature extraction via a neural network is shown in one embodiment of the present invention.

[0047] Figure 5 The diagram illustrates the processing steps of a robust transfer subspace learning algorithm in one embodiment of the present invention. Detailed Implementation

[0048] It should be noted that the components in the various figures may be shown exaggeratedly for illustrative purposes and are not necessarily to scale. In each figure, the same reference numerals are used for components that are identical or have the same function.

[0049] In this invention, unless otherwise specified, "arranged on," "arranged above," and "arranged on" do not exclude the possibility of an intermediate element between them. Furthermore, "arranged on or above" merely indicates the relative positional relationship between two components, and in certain cases, such as when the product orientation is reversed, it can also be converted to "arranged below or under," and vice versa.

[0050] In this invention, the various embodiments are merely intended to illustrate the solutions of the invention and should not be construed as limiting.

[0051] In this invention, unless otherwise specified, the quantifiers “a” and “one” do not exclude scenarios involving multiple elements.

[0052] It should also be noted that, in the embodiments of the present invention, only a portion of the components or parts may be shown for clarity and simplicity. However, those skilled in the art will understand that, under the teachings of the present invention, necessary components or parts can be added as needed for specific scenarios. Furthermore, unless otherwise stated, features in different embodiments of the present invention can be combined with each other. For example, a feature in the second embodiment can replace a corresponding or functionally identical or similar feature in the first embodiment, and the resulting embodiment will also fall within the scope of disclosure or description of this application.

[0053] It should also be noted that, within the scope of this invention, the terms "same," "equal," and "equal to" do not imply that the two values ​​are absolutely equal, but rather allow for a certain reasonable margin of error. In other words, the terms also encompass "substantially the same," "substantially equal," and "substantially equal to." Similarly, in this invention, the directional terms "perpendicular to," "parallel to," etc., also encompass the meanings of "substantially perpendicular to" and "substantially parallel to."

[0054] Furthermore, the numbering of the steps in the methods of the present invention does not limit the execution order of the method steps. Unless otherwise specified, the method steps may be executed in different orders.

[0055] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0056] Figure 1 A computer system implementing the system and / or method according to the present invention is shown. Unless otherwise specified, the method and / or system according to the present invention can be implemented in... Figure 1 The invention may be implemented in the computer system 100 shown to achieve the objectives of the invention, or the invention may be implemented in a distributed manner in multiple computer systems 100 according to the invention via a network, such as a local area network or the Internet. The computer system 100 of the invention may include various types of computer systems, such as handheld devices, laptop computers, personal digital assistants (PDAs), multiprocessor systems, microprocessor-based or programmable consumer electronic devices, network PCs, minicomputers, mainframes, network servers, tablet computers, etc.

[0057] like Figure 1 As shown, the computer system 100 includes a processor 111, a system bus 101, a system memory 102, a video adapter 105, an audio adapter 107, a hard disk drive interface 109, an optical drive interface 113, a network interface 114, and a Universal Serial Bus (USB) interface 112. The system bus 101 can be any of several bus architecture types, such as a memory bus or memory controller, a peripheral bus, and a local bus using various bus architectures. The system bus 101 is used for communication between the various bus devices. In addition to... Figure 1Other bus devices or interfaces besides those shown are also conceivable. System memory 102 includes read-only memory (ROM) 103 and random access memory (RAM) 104, wherein ROM 103 may, for example, store basic input / output system (BIOS) data for basic routines used to implement information transfer at startup, while RAM 104 provides the system with faster access speed running memory. Computer system 100 also includes hard disk drive 109 for reading and writing to hard disk 110, optical drive interface 113 for reading and writing to optical media such as CD-ROM, etc. Hard disk 110 may, for example, store operating system and application programs. Drives and their associated computer-readable media provide non-volatile storage for computer-readable instructions, data structures, program modules, and other data for computer system 100. Computer system 100 may also include a video adapter 105 for image processing and / or image output, which is used to connect output devices such as display 106. The computer system 100 may also include an audio adapter 107 for audio processing and / or audio output, which is used to connect output devices such as speakers 108. Furthermore, the computer system 100 may also include a network interface 114 for network connectivity, wherein the network interface 114 can be connected to the Internet 116 via a network device such as a router 115, wherein the connection can be wired or wireless. Additionally, the computer system 100 may also include a Universal Serial Bus (USB) interface 112 for connecting peripheral devices, wherein the peripheral devices include, for example, a keyboard 117, a mouse 118, and other peripheral devices such as microphones, cameras, etc.

[0058] When the present invention is Figure 1 When implemented on the computer system 100, the one-dimensional vibration signal can be converted into a two-dimensional image signal. The image signal allows for intuitive differentiation of signal differences between different fault types, providing a direct observation method. Furthermore, feature extraction is performed using a trained neural network. When facing unfamiliar operating conditions, only minor adjustments to the neural network are needed to continue using the model, significantly enhancing its robustness and reducing training costs. The system also considers the impact of Gaussian and non-Gaussian noise on the data, greatly improving the algorithm's stability. The use of a relaxation regression algorithm provides greater freedom for matrix transformations, which helps maintain the label structure during subspace transformations.

[0059] Furthermore, the embodiments can be provided as computer program products that may include one or more machine-readable media on which machine-executable instructions are stored, which, when executed by one or more machines such as a computer, computer network, or other electronic equipment, may cause one or more machines to perform operations according to the embodiments of the present invention. Machine-readable media may include, but are not limited to, floppy disks, optical disks, CD-ROMs (compact disc read-only memory) and magneto-optical disks, ROMs (read-only memory), RAMs (random access memory), EPROMs (erasable programmable read-only memory), EEPROMs (electrically erasable programmable read-only memory), magnetic or optical cards, flash memory, or other types of media / machine-readable media suitable for storing machine-executable instructions.

[0060] Furthermore, various embodiments can be downloaded as computer program products, wherein the program can be transmitted from a remote computer (e.g., a server) to a requesting computer (e.g., a client) via a communication link (e.g., a modem and / or a network connection) using one or more data signals implemented and / or modulated by a carrier wave or other propagation medium. Therefore, the machine-readable medium used herein may include such a carrier wave, but this is not required.

[0061] In this invention, the modules of the system according to the invention can be implemented using software, hardware, firmware, or a combination thereof. When a module is implemented using software, its function can be implemented through computer program flow. For example, the module can be implemented using code segments (such as code segments in languages ​​like C and C++) stored in a storage device (such as a hard disk, memory, etc.), wherein the corresponding function of the module can be implemented when the code segment is executed by a processor. When a module is implemented using hardware, its function can be implemented by setting a corresponding hardware structure. For example, the module's function can be implemented by hardware programming a programmable device such as a field-programmable gate array (FPGA), or by designing an application-specific integrated circuit (ASIC) that includes multiple transistors, resistors, capacitors, and other electronic devices. When a module is implemented using firmware, the module's function can be written into a read-only memory such as an EPROM or EEPROM in the form of program code, and the corresponding function of the module can be implemented when the program code is executed by a processor. In addition, some functions of the module may need to be implemented by separate hardware or by working in cooperation with the hardware. For example, the detection function is implemented by a corresponding sensor (such as a proximity sensor, accelerometer, gyroscope, etc.), the signal transmission function is implemented by a corresponding communication device (such as a Bluetooth device, infrared communication device, baseband communication device, Wi-Fi communication device, etc.), the output function is implemented by a corresponding output device (such as a display, speaker, etc.), and so on.

[0062] Figure 2A flowchart illustrating a fault diagnosis method based on robust transfer subspace learning in one embodiment of the present invention is shown. Figure 2 As shown, the method may include the following steps:

[0063] Step 201: Convert the one-dimensional vibration signal into a two-dimensional image.

[0064] Step 202: Provide a neural network and extract features from a two-dimensional image using the neural network.

[0065] Step 203: Perform transfer learning based on the extracted features using a robust transfer subspace learning algorithm.

[0066] Step 204: Determine the fault type corresponding to the one-dimensional vibration signal based on the results of transfer learning.

[0067] The steps of this method are further explained below with reference to the accompanying drawings.

[0068] This method first requires dividing the transfer learning dataset according to different operating conditions (mainly different load speeds).

[0069] Furthermore, it is necessary to convert the one-dimensional vibration signal into a two-dimensional grayscale image. Figure 3 This illustration shows a process for converting a one-dimensional signal into a two-dimensional image according to an embodiment of the present invention. The image size and number of images are determined based on the size of the vibration signal dataset. Taking the Case Western Reserve University dataset (CWRU) as an example, this dataset includes hundreds of thousands of vibration signals. The corresponding two-dimensional grayscale image can have 128×128 pixels and 10 images. The corresponding number of signal points is 128×128×10. The number of signal points is close to the number of vibration signals, which avoids repeated coverage of the signals by the images.

[0070] The process of converting a one-dimensional vibration signal into a two-dimensional grayscale image can be represented by the following formula:

[0071]

[0072] Where P(j,k) represents a pixel in the image matrix, j and k represent the row and column of the image matrix, respectively, L represents a signal point segment, and L((j-1)×M+k) represents the vibration signal. L can be a segment of 128×128 signal points. This formula can convert 128×128 vibration signals into 128×128 grayscale values ​​between 0 and 255, and the resulting grayscale values ​​can be used as pixels.

[0073] Furthermore, features are extracted from the converted 2D grayscale image using a neural network. This can be achieved by training the neural network on an image dataset (e.g., using a VGG-19 convolutional neural network), extracting features through the trained network, and outputting 4096-dimensional features in the fully connected layers of the neural network. Figure 4 A schematic diagram illustrating image feature extraction via a neural network is shown in one embodiment of the present invention. Figure 4 As shown, in the neural network, a two-dimensional image is passed through a convolutional layer (Conv+ReLU) 401, a pooling layer (Max Pooling) 402, a fully connected layer (Fully Connected+ReLU) 403, and a normalized exponential layer (Softmax) 404 to generate extracted features.

[0074] Through the above steps, each two-dimensional grayscale image can be reduced to a 4096-dimensional feature. The source and target domain datasets of the neural network are transformed into s×4096 and t×4096 matrices, respectively, where s and t represent the number of images in the source and target domain datasets, respectively. The resulting matrices represent the feature matrices of the source and target domains, respectively.

[0075] Furthermore, a robust transfer subspace learning algorithm is used to perform transfer learning on the feature matrices of the source domain and the feature matrices of the target domain. Figure 5 The diagram illustrates the processing procedure of a robust transfer subspace learning algorithm according to an embodiment of the present invention. The algorithm iteratively generates a transformation matrix, which transforms the source and target domains into a common subspace. Within this common subspace, the differences between the source and target domains (i.e., the differences between the feature matrices of the source and target domains) can be reduced, thereby enabling transfer learning.

[0076] The robust transfer subspace learning algorithm can find a linear transformation matrix P, which can transform the source and target domains. After the transformation, the target domain data can be linearly represented by the source domain data, where Z represents the reconstruction matrix. During the transformation process, a 1-norm and a kernel norm can be applied to the reconstruction matrix Z to preserve the local and global structure of the data.

[0077] Furthermore, in the robust transfer subspace learning algorithm, a non-Gaussian noise matrix E and a Gaussian noise matrix N can be introduced to reduce the impact of both types of noise on the data. A relaxation regression algorithm can also be introduced to preserve the label structure of the samples. The relaxation regression algorithm can directly learn the labels from the input matrix, thus providing higher degrees of freedom to the transformation matrix P. The robust transfer subspace learning algorithm can be expressed as follows:

[0078]

[0079] in, The function represents the relaxation regression algorithm, R represents the learned label matrix, Z represents the reconstruction matrix, E represents the non-Gaussian noise matrix, N represents the Gaussian noise matrix, and ||·|| * The nuclear norm is represented by ||·||1, which represents the 1-norm. F Let F be the norm, and α, β, γ represent the nonnegativity parameters. i This represents the index of the true label of the i-th sample in the source domain.

[0080] This method transforms one-dimensional vibration signals into two-dimensional image signals, allowing for intuitive differentiation of signal differences between different fault types and providing a direct observation method. The method extracts features using a trained neural network, enabling continued use under unfamiliar operating conditions with only minor adjustments to the neural network, significantly enhancing model robustness and reducing training costs. Furthermore, this method considers the impact of Gaussian and non-Gaussian noise on the data, greatly improving algorithm stability. The use of relaxation regression provides greater freedom for matrix transformations, helping to preserve the label structure during subspace transformations.

[0081] In one embodiment of the present invention, using this method to perform unsupervised fault diagnosis under unknown operating conditions may include the following steps:

[0082] The one-dimensional source domain vibration signal data S1 and the one-dimensional target domain vibration signal data S2 are transformed to obtain two-dimensional image data of the vibration signal, represented as X1 and X2, respectively, where the source domain label is known as Y1. X1 is used as the input of a VGG-19 convolutional neural network for training. The model is saved when it converges and the classification accuracy is high enough.

[0083] The parameters of the model are frozen and saved. X1 and X2 are used as inputs to the model. The 4096-dimensional features of the image are output in the fully connected layer of the frozen model. After feature extraction, the feature matrices of the source domain and target domain vibration signal images are obtained, denoted as Z1 and Z2, respectively. The size of the feature matrices is s×4096 and t×4096, where s and t are the number of source domain and target domain images, respectively.

[0084] Using two feature matrices as input to a robust transition subspace learning algorithm, Z1 and Z2 are transformed into a common subspace. The algorithm eventually obtains an optimal transformation matrix P through multiple iterations. During the transformation process, Gaussian and non-Gaussian noise are removed, and a relaxation regression algorithm is used to increase the degrees of freedom of the common subspace.

[0085] Multiplying the transformation matrix P with the original matrices Z1 and Z2 yields the source and target domain data M1 and M2 in the common subspace. M1 and Y1 are then used as inputs to a KNN classifier for training until the classifier converges. M2 is then used as input to the classifier to predict the fault type of the vibration signal in the target domain.

[0086] Although various embodiments of the invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined solely by the appended claims and their equivalents.

Claims

1. A fault diagnosis method based on robust transfer subspace learning, characterized in that, comprising the following steps: transforming the one-dimensional vibration signal into a two-dimensional image; providing a neural network and extracting features from the two-dimensional image by the neural network; performing transfer learning based on a robust transfer subspace learning algorithm according to the extracted features, comprising: based on robust transfer subspace learning algorithm, the source domain feature matrix and the target domain feature matrix are transformed into a common subspace, wherein the transformation matrix is generated by iteration , the Gaussian noise and the non-Gaussian noise are eliminated during the transformation process, and the freedom degree of the common subspace is improved by a relaxation regression algorithm; transforming the matrix with the source domain feature matrix to generate a common subspace source domain matrix ; and transforming the matrix with the target domain feature matrix to generate a common subspace target domain matrix ; and determining a fault type corresponding to the one-dimensional vibration signal according to a result of the transfer learning, comprising: with the public subspace source domain matrix and source domain labels train a KNN classifier as input; and by the trained KNN classifier according to the common subspace target domain matrix determining a one-dimensional target domain vibration signal corresponding fault type. 2.The robust shift subspace learning based fault diagnosis method according to claim 1, characterized in that, transforming the one-dimensional vibration signal into a two-dimensional image comprises: generating a two-dimensional image matrix according to the one-dimensional vibration signal, denoted as follows: wherein, represents a pixel of a two-dimensional image matrix, respectively represent a row and a column of the image matrix, represents a signal point paragraph, represents a one-dimensional vibration signal, represents an image gray value range, wherein the image gray value range is 0-255. 3.The robust shift-subspace learning based fault diagnosis method according to claim 2, characterized in that, transforming the one-dimensional vibration signal into a two-dimensional image comprises: transforming one-dimensional source domain vibration signals into two-dimensional source domain images , wherein the source domain labels of the two-dimensional source domain images are known; and transforming one-dimensional target domain vibration signals into two-dimensional target domain images . 4.The robust shift-subspace learning based fault diagnosis method of claim 2, wherein, providing a neural network and extracting features from the two-dimensional image by the neural network comprises: with the two-dimensional source domain image training a neural network as input; from the two-dimensional source domain images by the trained neural network extracting a source domain feature matrix wherein denotes the number of two-dimensional source domain images; and extracting a target domain feature matrix from a two-dimensional target domain image by the trained neural network wherein the number of two-dimensional target domain images is represented by N.

5. The robust shift subspace learning based fault diagnosis method according to claim 4, characterized in that, the neural network comprises a convolution layer, a pooling layer, a fully connected layer, and a normalized exponential layer. 6.The robust shift-subspace learning based fault diagnosis method of claim 5, wherein, the improving the degree of freedom of the common subspace by the relaxation regression algorithm is denoted as follows: wherein, represents a relaxed regression algorithm, represents a learned label matrix, represents a reconstruction matrix, represents a non-Gaussian noise matrix, represents a Gaussian noise matrix, represents a kernel norm, represents a 1-norm, represents norm, represents a non-negative parameter, represents an index of a true label of a th sample in the source domain; transforming the matrix with the source domain feature matrix to generate a common subspace source domain matrix ; and transforming the matrix with the target domain feature matrix to generate a common subspace target domain matrix .

7. A fault diagnosis system based on robust transfer subspace learning, characterized in that, comprising: an image transformation module configured to transform the one-dimensional vibration signal into a two-dimensional image; a feature extraction module configured to train a neural network and extract features from the two-dimensional image by the neural network; a transfer learning module configured to perform transfer learning based on a robust transfer subspace learning algorithm according to the extracted features, the performing transfer learning based on a robust transfer subspace learning algorithm according to the extracted features comprising: based on robust transfer subspace learning algorithm, the source domain feature matrix and the target domain feature matrix are transformed into a common subspace, wherein the transformation matrix is generated by iteration, the Gaussian noise and the non-Gaussian noise are eliminated during the transformation process, and the freedom of the common subspace is improved by a relaxation regression algorithm; transforming the matrix with the source domain feature matrix to generate a common subspace source domain matrix ; and transforming the matrix with the target domain feature matrix to generate a common subspace target domain matrix ; and a fault prediction module configured to determine a fault type corresponding to the one-dimensional vibration signal according to a result of the transfer learning, the determining a fault type corresponding to the one-dimensional vibration signal according to a result of the transfer learning comprising: with the public subspace source domain matrix and source domain labels train a KNN classifier as input; and by the trained KNN classifier according to the common subspace target domain matrix determining a one-dimensional target domain vibration signal corresponding fault type.

8. A computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method according to any one of claims 1-6.

9. A computer system comprising: a processor configured to execute machine executable instructions; and a memory having stored thereon machine executable instructions which, when executed by the processor, perform the steps of the method according to any one of claims 1-6. ​

Images